Operational intelligence · Data leadership · AI transformation

I turn fragmented operations into systems that compound.

Eight years building the operational-intelligence layer of a $100M business: forecasting, costing, replenishment, reporting. Most recently AI-native: 4,200 hours of recurring work automated in seven months. The pattern is the same every time: find the leverage, build the thing, work myself out of the job.

8 years

building the operational-intelligence layer of a $100M business. Everything else on this page is detail under that.

Scroll for the work Calgary tyler@tylerdove.me

How I work

Operations, built as systems, not headcount.

The work that keeps a business running (planning, forecasting, costing, replenishment, reporting) is operations. I build it as systems and automation, not headcount.

Find

The decisions that cost money when they're slow.

Most operations bleed value in the same place: a decision that could be made today gets made next month. I go find those: the points where one hour of system work saves a hundred hours of human work.

Fix

Map back from the decision to its inputs.

Once you know which decision matters, you work backward: what information would make it faster, where that information lives, and what has to be built or automated to put it in front of the right person at the right time.

Prove

Done when the decision got faster.

A project isn't done when it's built; it's done when the decision it serves got faster or better. Same loop whether I'm building it hands-on or standing up the function.

Selected work

Selected work.

Two pillars at depth (the foundational build that establishes the operator, then the AI mandate that sits on top) and a short turnaround. The numbers live in the results.

Pillar 01
$100M live-plant wholesaler
2016–2022 · Senior data analyst & leadership team

Building the data department

From one sentence to the company's full operational-intelligence function.

The mandate

Built the data department from scratch and joined the leadership team. The original mandate was one sentence: "get the right product to the right store at the right time, every time", and I grew it into the company's full operational-intelligence and operations function.

What I built

Four systems, each one a class-solver rather than a one-off: a just-in-time replenishment model, a full cost-and-pricing model, a warehouse management system, and a nursery inventory system. Built hands-on in Excel and SQL, iterated for years across changing product lines, customers, and seasonality.

The result

100,000+ forecasts a week across 1,000+ stores and 2,000+ perishable SKUs. One retailer went 17th to 2nd in national grossing rank in a single year after ceding inventory control to the model, with lifts of 20% or more typical.

Stack
Excel, deepSQLPostgreSQLRelational data modelsForecasting at scale

"Get the right product to the right store at the right time, every time."

The entire job description. One letter-sized page, one sentence.

One customer's national grossing rank, one year
Before the model managed inventory
17th
After
1st
2nd
3rd

Illustrative bars, real ranks. Typical lift when stores ceded control: 20% or more.

Four systems,
built from scratch
A

Replenishment model

100,000+ forecasts a week. 17th to 2nd in a single year.

Built the just-in-time replenishment model from scratch in Excel, then migrated it to SQL once it outgrew the spreadsheet. It translates a 14-day SKU-by-store sell-through forecast into specific shipment quantities, accounting for supply constraints, store-level trends, and minimum economic order quantities. Iterated for five-plus years across changing product lines, customers, and seasonality. Award-winning; proprietary, so no live demo. That's the honest answer.

Excel, deepSQLPostgreSQL
Right product · right store · right time
A 14-day forecast becomes a shipment
SKU×store
Supply
Trend
Min EOQ

100,000+ unique forecasts per week, $100M+/yr of product movement on daily recommendations.

B

Costing & pricing model

Full cost build across 2,000+ SKUs.

A full cost build across 2,000+ SKUs: direct labour, materials, and direct and indirect overhead, allocated down to the unit. Sales used it to steer customers toward margin, pricing from a defensible cost floor instead of a gut feel. The same relational thinking that powered replenishment, pointed at the P&L.

Excel, deepCost accountingSQL
2,000+ SKUs costed

direct labour, materials, direct + indirect overhead: every line traceable to a margin.

Sales priced from the cost floor, not a gut feel.

C

Warehouse management system

One spreadsheet on a thumb drive, replaced. 20M+ units tracked.

Inventory ran on a single spreadsheet on a thumb drive. I replaced it with a full PO, receiving, inventory, and transfer system tied directly to production, tracking 20M+ units. It drew down aging stock and cut production stoppages: the operational floor under everything the forecasts assumed.

SQLPO / receiving / transfersProduction integration
Before
  • One spreadsheet on a thumb drive
  • No live receiving or transfers
  • Aging stock, invisible
  • Production stoppages
After
  • PO, receiving, inventory, transfers
  • Tied to production, 20M+ units
  • Aging stock drawn down
  • Stoppages cut

A real system where there had been a file, tracking 20M+ units against production.

D

Nursery inventory system

600 acres, 500+ varieties, grow times up to 5 years.

Managed 600 acres and 500+ varieties with grow times stretching up to five years, optimizing what to grow as a plant's value shifts by age and stage. A forecasting problem in a different shape: the inventory is alive, the lead time is measured in seasons, and the wrong call compounds for years.

Excel, deepSQLMulti-year planning
600 acres · 5-year horizon

500+ varieties, optimized by age and stage: value shifts as a plant matures.

The inventory is alive; the wrong call compounds for years.

Pillar 02
$100M live-plant wholesaler
2025–26 contract · AI lead

The AI integration mandate

4,200 hours of human work, automated in seven months.

The problem

Leadership put it bluntly: "We've heard about AI for years. We don't know what to do with it." No roadmap, no use cases, no internal expertise. Just a mandate and a number to hit.

What I did

Ran discovery, then built. Used n8n as the orchestration layer and Claude as the language layer. Tried more than 30 workflow concepts and shipped 12 to production. The undocumented spaghetti at the wall was deliberate: you don't know which automation earns its keep until you build the cheap version and watch it run.

The result

The equivalent of 4,200 human hours per year automated, against a 4,000-hour target, in roughly seven months. The shipped workflows are the systems below.

Stack
n8nClaudeZoho CRM v8 APIOutlook + Microsoft GraphInternal sales, credit, shipment DBsGoogle SheetsConfidence scoringScripted + LLM-as-user evals
12 shipped to production 18 retired early, cheaply

Every concept got the cheap version first. The 18 that did not earn their keep cost almost nothing to kill.

We've heard about AI for years. We don't know what to do with it." I told them I didn't either, but I'd figure it out. The brief from leadership, and the answer
The 12 shipped,
three that matter most
A

Customer Intelligence Brief

Five thousand fragmented contact histories, one 200-word brief each.

History was scattered across a CRM, years of email, and internal sales, credit, and shipment databases. New staff called people knowing nothing, and annual turnover walked the institutional memory out the door. This workflow aggregates everything into a single one-page brief per contact, paired with an Email-to-Notes capture system that records outbound communications back into the knowledge base. 5,000+ contacts covered. Better notes feed better briefs, which carry customer continuity through staff turnover.

n8nClaudeZoho CRM v8Outlook + Graph
Contact Brief 200 words, 1 page
Relationship
History, sales + credit + shipments
Open items
Conversational landmines

Structure of the real brief, content redacted. One of these exists for 5,000+ contacts.

B

SMART Goals Chat Assistant

A three-hour task, done in five minutes. Highest adoption in company history.

Employees dreaded writing SMART goals for performance reviews. Some spent up to three hours. Most avoided it. The assistant interviews the employee and walks them through writing a goal, turning a blank-page problem into a guided five-minute chat. About 100 employees adopted it in a single review cycle, roughly 140 to 150 unique runs: the highest adoption rate of any internal tool in the company's history. Tested with scripted conversation evals and LLM-as-user simulation evals.

n8nClaudeScripted evalsLLM-as-user evals
What part of your work would feel different if next quarter went well?
Honestly, month-end close. It always runs late.
Late by how much, and what usually causes it?
Two or three days. Waiting on branch numbers, mostly.
Then here is a draft to react to: close the books within five business days of month-end, every month this quarter, by moving branch submissions to a fixed Tuesday deadline. Measurable, time-bound, yours. Want to tighten anything?

Reconstruction with a throwaway example, not employee data.

C

The accountability layer

Priorities made visible without adding a single meeting.

Project status lived in people's heads, and surfacing it meant more meetings. So: a set of async automations instead. The PM Assistant sends portfolio managers a weekly brief on every project in their book, plus twice-weekly nudges to owners (Monday, here are this week's three; Thursday, no progress logged, what's up). Store Pulse produces a Friday 3-2-1 chain summary. A Portfolio Digest rolls it all up for leadership.

n8nClaudeHTML email
Store Pulse · Friday 3-2-1
3 updates
2 decisions needed
1 ask

The actual email format, content redacted. Lands every Friday without anyone booking a room.

Also
Former employer
Turnaround lead

The turnaround

A leaderless data team, 200+ deliverables, mapped and handed back thriving.

The situation

Came back to a former employer as a contractor when the person who replaced me left. The data team was running without weekly direction.

What I did

Mapped 200+ specific deliverables across the year, assigned and resourced them, and built a dashboard to monitor execution. Ran the team day to day, including performance evaluations of people who used to be peers. Rebuilt the replenishment model for half the business, rebuilt the nursery distribution allocation system from scratch, and hired and trained the permanent department lead.

Stack
ExcelSQLPower BIDashboards
0+ deliverables

mapped, assigned, resourced, and monitored on one dashboard.

Handed back thriving

a three-person team with documented processes and a permanent leader in the seat. The point of the job was to stop being needed.

Additional builds

Real things, built AI-native.

Beyond the core roles: full-stack web apps and agentic systems, shipped with Claude, not written from scratch.

In progress

Pacific Agriculture web app

Digitized a crop-insurance inspection workflow that had run on Drive folders, spreadsheets, iPhone Notes, and paper. Spec-first (a requirements traceability matrix, ~150 features, a decision register), then built: a relational CRM, a file-to-submit claim workflow, an EXIF-to-map photo pipeline, and ministry-PDF auto-generation. A foundation that compounds as more data flows through it.

Next.jsConvexBuilt AI-native with Claude Code
accidentalpm.ai

Accidental PM

A conversational AI project-management assistant: it scopes plans, tracks blockers, and drafts the nudges that keep work moving: the accountability layer, productized.

Next.jsClauden8n
Research only, no live trading

Agentic Quant-Trading Researcher

An agentic system that takes a strategy, backtests it, critiques the result, and modifies it iteratively. Built as separate Scanner, Risk, and Journal agents.

Python, shipped with ClaudeSQLNode.js

Skills and credentials

Stated honestly. A visible gap beats vague language.

Deep
  • Excel and financial modeling, ~10,000 hours, usually the best in the room
  • Forecasting at scale, 100K+ forecasts per week
  • AI workflows with LLMs in production
  • Aggregating fragmented data into actionable briefs
  • Relational data models, designed from scratch
  • Process design
  • SQL, PostgreSQL, and relational thinking
Working
  • n8n, deep
  • Next.js, React, Convex, Supabase, with AI assist
  • Zoho CRM, Outlook and Microsoft Graph
  • Advanced Google Sheets, Airtable
  • D3.js, SVG, Canvas
  • Claude and Claude Code as daily drivers
  • Python, shipped with Claude, not written from scratch
Facilitation and method
  • Full Stack Facilitation, AJ&Smart certified
  • Design Thinking
  • Adizes EI Council
  • Systems thinking
  • Process improvement
Credentials
Full Stack Facilitator, AJ&Smart Palo Alto cohort, 2024 AJ&Smart is the firm behind the Workshopper and Design Sprint methodology, widely considered the leading facilitation training in the world. One of, possibly the only, certified Full Stack Facilitators in Canada.
Design Thinking, BrainStation Multi-day intensive
BBA (Honours), University of the Fraser Valley 2016 Double major in Finance and Accounting, minor in Economics.

The throughline

It's the same job in three disguises.

I started in finance and accounting, spent most of my career as the data person (building a department from nothing at a $100M wholesaler and iterating an award-winning forecasting model for years), and the last while has been about AI. The throughline is the same across all of it: I find the points where one hour of system work saves a hundred hours of human work, build the thing, and work myself out of the job.

Tyler Dove is a Calgary-based operator. He spent eight years building the operational-intelligence layer of a $100M live-plant wholesaler, where he built the data department from scratch and sat on the leadership team.

Independent since 2022, across data leadership, AI builds, full-stack web apps, and executive facilitation, usually two to four engagements running at once. A certified Full Stack Facilitator who works AI-native, living in Claude Code and shipping through n8n.

Contact

Bring me in to build it hands-on, or to stand up and lead the function you've already started. Same loop either way.

tyler@tylerdove.me

Based in Calgary, remote-native. Every number on this page is defensible in a reference check, and I'd rather ship a rough v1 this week than wait for a perfect rollout. LinkedIn.